вход по аккаунту



код для вставкиСкачать
Tussle in Cyberspace: Defining Tomorrow’s Internet
David D. Clark
John Wroclawski
MIT Lab for Computer Science
MIT Lab for Computer Science
Karen R. Sollins
Robert Braden
MIT Lab for Computer Science
USC Information Sciences Institute
The architecture of the Internet is based on a number of
principles, including the self-describing datagram packet,
the end to end arguments, diversity in technology and global
addressing. As the Internet has moved from a research curiosity to a recognized component of mainstream society,
new requirements have emerged that suggest new design
principles, and perhaps suggest that we revisit some old
ones. This paper explores one important reality that surrounds the Internet today: different stakeholders that are
part of the Internet milieu have interests that may be adverse to each other, and these parties each vie to favor their
particular interests. We call this process “the tussle”. Our
position is that accommodating this tussle is crucial to the
evolution of the network’s technical architecture. We discuss some examples of tussle, and offer some technical design
principles that take it into account.
The Internet was created in simpler times. Its creators
and early users shared a common goal—they wanted to build
a network infrastructure to hook all the computers in the
world together so that as yet unknown applications could
be invented to run there. All the players, whether designers,
users or operators, shared a consistent vision and a common
sense of purpose.
Perhaps the most important consequence of the Internet’s
success is that the common purpose that launched and nurtured it no longer prevails. There are, and have been for
some time, important and powerful players that make up
the Internet milieu with interests directly at odds with each
Some examples are very current. Music lovers of a certain bent want to exchange recordings with each other, but
the rights holders want to stop them. People want to talk
in private, and the government wants to tap their conversations. Some examples are so obvious that they are almost
overlooked. For the Internet to provide universal interconnection, ISPs must interconnect, but ISPs are sometimes
fierce competitors. It is not at all clear what interests are
being served, to whose advantage, to what degree, when
ISPs negotiate terms of connection. It is not a single happy
family of people dedicated to universal packet carriage.
We suggest that this development imposes new requirements on the Internet’s technical architecture. These new
requirements, in turn, motivate new design strategies to accommodate the growing tussle among and between different
Internet players. The purpose of this paper is to explore
what these requirements and strategies might be.
We begin by briefly discussing the Internet landscape some fundamental differences between the mechanisms of
engineering and society, and the players that populate our
field. We then outline some proposed design principles intended to accommodate within the Internet mechanisms of
society as well as those of engineering. We believe this accommodation is central to designing an Internet that is resilient to the challenges of society as well as those of technology. We conclude by discussing some tussle spaces, ways
in which our principles might guide the technical response
to these spaces, and specific technical research that may be
of value in accommodating these tussles.
Categories and Subject Descriptors
C.2.1 [Computer Systems Organization]: Computer
Communications Networks—Network Architecture and Design; H.1 [Information Systems]: Models and Principles;
K.4.1 [Computing Milieux]: Computers and Society—
Public Policy Issues
General Terms
Design, Economics, Legal Aspects, Security, Standardization
Tussle, Network Architecture, Trust, Economics, Design
Principles, Competition
Work sponsored in part by the Defense Advanced Research Projects Agency
(DARPA) and Air Force Research Laboratory, Air Force Materiel Command, USAF, under agreement number F30602-00-2-0553 at MIT, and
agreement number F30602-00-1-0540 at ISI. The U.S. Government is authorized to reproduce and distribute reprints for Governmental purposes
notwithstanding any copyright annotation thereon.
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. To copy otherwise, to
republish, to post on servers or to redistribute to lists, requires prior specific
permission and/or a fee.
SIGCOMM’02, August 19-23, 2002, Pittsburgh, Pennsylvania, USA.
Copyright 2002 ACM 1-58113-570-X/02/0008 ...$5.00.
The natures of engineering and society
Engineers attempt to solve problems by designing mech-
There is great diversity within each of these categories:
there are “good users” and spammers, dominant ISPs and
small players, private providers with more or less rigidity
about usage, liberal and conservative governments, and so
on. The resulting tussles span a broad scope: the rights
of the individual vs. the state, the profit seeking of competitors, the resistance to those with malicious intent, those
with secrets vs. those who would reveal them, and those
who seek anonymity vs. those would identify them and hold
them accountable. The list probably mirrors every aspect
of society. For a detailed discussion of these various players
and their impact on the Internet, see [1].
anisms with predictable consequences. Successful engineering yields bridges that predictably don’t fall down, planes
that predictably don’t fall out of the sky, and calculators
that give the “right” answer. The essence of engineering is
the development and codification of models, techniques and
tools that deliver predictable, desirable behavior.
The technical development of the Internet has followed
this path. As a community, we focus on design principles that deliver such virtues as robustness, scalability and
manageability in the face of complexity, component failures,
growth, and other challenges. However, as the Internet becomes mainstream it inevitably moves from being an engineering curiosity to being a mirror of the societies in which
it operates. The Internet may have been designed by engineers, but its behavior (and its evolution) is by no means
predictable today.
The operation of societies follows a different model. Historically, the essence of successful societies is the dynamic
management of evolving and conflicting interests. Such societies are structured around ‘controlled tussle’ – regulated
by mechanisms such as laws, judges, societal opinion, shared
values, and the like. Today, this is the way the Internet is
defined—by a series of ongoing tussles. Different parties
adapt its mix of mechanisms to try to achieve their conflicting goals, and others respond by adapting the mechanisms
to push back. Thus, conservative governments and corporations put their users behind firewalls, and the users route
and tunnel around them. ISPs give their users a single IP
address, and users attach a network of computers using address translation. There is no “final outcome” of these interactions, no stable point, and no acquiescence to a static
architectural model.
The challenge facing Internet research and engineering is
to recognize and leverage this reality – at minimum to accommodate it; if possible, to use it to strengthen the technical architecture. In other words, the technical architecture
must accommodate the tussles of society, while continuing
to achieve its traditional goals of scalability, reliability, and
evolvability. This expansion of the Internet’s architectural
goals is a difficult, but central technical problem.
The thesis of this paper is that the future of the Internet will increasingly be defined by tussles that arise among
the various parties with divergent interests, and that the
technical architecture of the Internet must respond to this
observation. If this is so, are there principles to guide designers, and mechanisms that we should use in recognition
of this fact?
In this paper we offer some design principles to deal with
tussle. Our highest-level principle is:
• Design for variation in outcome, so that the outcome
can be different in different places, and the tussle takes
place within the design, not by distorting or violating
it. Do not design so as to dictate the outcome. Rigid
designs will be broken; designs that permit variation
will flex under pressure and survive.
Within this guiding principle, we identify two more specific
• Modularize the design along tussle boundaries, so that
one tussle does not spill over and distort unrelated
• Design for choice, to permit the different players to
express their preferences.
The Internet landscape
Modularize along tussle boundaries
Systems designers know to break complex systems into
modular parts. Modularity is typically used to manage complexity, allow for independent implementation and component reuse, or to meet other technical goals. But “tussle
isolation” is perhaps a new principle.
Today, there are many parties that are part of the Internet
milieu. These include:
• Users, who want to run applications and interact over
the Internet.
• Functions that are within a tussle space should be logically separated from functions outside of that space,
even if there is no compelling technical reason to do so.
Doing this allows a tussle to be played out with minimal distortion of other aspects of the system’s function.
• Commercial ISPs, who sell Internet service with the
goal of profit.
• Private sector network providers who run a part of the
Internet to facilitate their business or other undertaking.
The design of the DNS provides an example. The current
design is entangled in debate because DNS names are used
both to name machines and to express trademark. In retrospect, one might have predicted that fights over trademarks
would be a tussle space, and made sure that the names that
expressed trademarks were used for as little else as possible.
In particular, one might imagine separate strategies to deal
with the issues of trademark, naming mailbox services, and
• Governments, who enforce laws, protect consumers,
regulate commerce, and so on.
• Intellectual property rights holders, who want to protect their materials on the Internet.
• Providers of content and higher level services, offered
in search of profit or as a public service.
Choice often requires open interfaces. Open interfaces have played a critical role in the evolution of the Internet, by allowing for competition among algorithms, implementations, and vendors, and by enabling rapid technical progress through replacement of modular parts rather
than entire systems. But open interfaces also allow choice,
not just replacement. If a protocol allows a party to select
among alternative providers of service, this usually implies
that the interface to that service is well-defined, so that independent versions of the service can be constructed.
Tussles often happen across interfaces. Some tussles
involve the use of different mechanisms by different parties.
However, some tussles, such as the tussle among competitive ISPs, may involve technical interfaces between those
parties. For example, BGP is used as the routing protocol
among ISPs, who interconnect but are business competitors.
If an interface occurs at a point of tussle, it will have different attributes from an interface that exists simply to foster
interoperability, modularity, or competition among suppliers.
Open interfaces at tussle points may benefit from the following sorts of properties, which are not always important
in other cases.
providing names for machines that are independent of location (the original and minimal purpose of the DNS). One
could then try to design these latter mechanisms to try to
duck the issue of trademark.
• Solutions that are less efficient from a technical perspective may do a better job of isolating the collateral
damage of tussle.
In contrast, the current trajectory of IP QoS design tries
to isolate tussles. The use of explicit ToS bits to select QoS,
rather than binding this decision to another property such
as a well-known port number, disentangles what application
is running from what service is desired. It can be anticipated
that there will be tussles about what applications a user can
run, and separately tussles about what service qualities can
be provided. The designers felt that it was better to separate these ideas. This modularity allows tussles about QoS
to be played out without distortions, such as demands that
encryption be avoided simply to leave well-known port information visible or the encapsulation of applications inside
other applications simply to receive better service.
Design for choice
• Visible exchange of value.
Network protocols are designed so that different parties on
the network can communicate with each other, consumers
can make use of the resources of providers, and providers can
interconnect with each other to provide service. It is important that protocols be designed in such a way that all the
parties to an interaction have the ability to express preference about which other parties they interact with. Protocols
must permit all the parties to express choice.
For example, the design of the mail system allows the user
to select his SMTP server and his POP server. A user can
pick among servers, perhaps to avoid an unreliable one or
pick one with desirable features, such as spam filters. Users
can select what news server they use, perhaps to prevent
their children from encountering some of the more colorful news groups. This sort of choice drives innovation and
product enhancement, and imposes discipline on the marketplace.
The form that the choice takes for the different parties
may be different. A user of mail might choose her SMTP
server by configuring a mail-sending program. An ISP might
try to control what SMTP server a customer uses by redirecting packets based on the port number.1
Providing this sort of choice has a drawback—it adds to
the complexity of configuring and using a service. For naı̈ve
users, choice may be a burden, not a blessing. To compensate for this complexity, we may see the emergence of third
parties that rate services (the on-line analog of Consumers
Reports) and parties that provide pre-configured software
to relieve the user of dealing with the details of choice.
• Exposure of cost of choice.
• Visibility (or not) of choices made.
• Tools to resolve and isolate faults and failures.
It matters if the consequence of choice is visible.
Choices made in public are sometimes different than those
made in secret. In some cases, there is no way to hide a
choice. Often, the choice can be secret, even if its consequences are visible. The routing arrangements among ISPs
are generally not public, even though everyone can see the
consequences at the BGP level. A link-state routing protocol requires that everyone export his link costs, while a
distance vector protocol makes it harder to see what the
internal choices are.
Tussles have different flavors. In some cases, the interests of the players are simply adverse, and there is no
win-win way to balance them. But in many cases, players’
interests are not adverse, but simply different. A user wants
to send data; a provider wants to be compensated for carrying it. While this implies a natural tussle over pricing, in
the end both parties realize that they must meet the other’s
To support this class of tussle, recognize that there is often
an exchange of value for service. Value need not be “money”
but often will be. Napster is a non-monetary example that
illustrates the “mutual aid” aspect of peer-to-peer networking. Whatever the compensation, recognize that it must
flow, just as much as data must flow. Sometimes this happens outside the system, sometimes within a protocol. If
this “value flow” requires a protocol, design it. (There is an
interesting case study in the rise and fall of micro-payments,
the success of the traditional credit card companies for Internet payments, and the emergence of PayPal and similar
Tussles evolve over time. A traditional engineering
design produces a result that is constant until the mechanism is redesigned. But tussle is ongoing and evolutionary.
These principles, and the reality of tussle, have some further implications for design:
An over-generalization of the tussle is that service providers
exercise control over routing; end-users control selection of
other end-points. End-users try to over-rule constrained
routing with tunnels and overlay networks.
ISPs would like to find ways to lock in their customers;
customers want to preserve the ability to change among
providers. This illustrates the basic consumer-producer tussle in a competitive world. For hosts that use static addresses, renumbering is a complex task. Because renumbering hosts can be hard, there is a very explicit tension
today between the desire to have addresses reflect topology
to support efficient routing and the desire of the customer to
change providers easily. Either a customer is locked into his
provider by the provider-based addresses, or they obtain a
separate block of addresses that are not topologically significant and therefore add to the size of the forwarding tables
in the core of the network. Many ISP’s refuse to route small
address blocks, nominally protecting the routing tables but
also locking the customer to their address range. The responses by the consumer include dynamic host numbering
(DHCP) and dynamic update of DNS entries when the host
is renumbered.
Each side finds new ways to gain an advantage, and then
the other side responds. This implies, first, that any thinking about tussle must view it as a multi-round process, and
second, that as mechanism is drawn into an ongoing tussle,
it may be used in unexpected ways, and require redesign to
survive in this new role.
There is no such thing as value-neutral design.
What choices designers include or exclude, what interfaces
are defined or not, what protocols are open or proprietary,
can have a profound influence on the shape of the Internet,
the motivations of the players, and the potential for distortion of the architecture.
Don’t assume that you design the answer. You are
designing a playing field, not the outcome.
In this section we discuss some specific aspects of the Internet in which different players with competing interests
come together. In each case, our goal is to examine the nature of the tussle and to illustrate how our principles can be
applied in specific cases. We suggest some specific research
areas that would benefit from application of our principles.
• A desire for vigorous competition would suggest that
the consumer should have the choice to move from ISP
to ISP. Given that, the Internet design should incorporate mechanisms that make it easy for a host to change
addresses and to have and use multiple addresses.
Addresses should reflect connectivity, not identity, to
modularize tussle. This would relieve problems with
end-node mobility, improve choice in multi-homed machines, and improve the ease of changing providers.
One of the tussles that define the current Internet is the
tussle of economics. The providers of the Internet are not in
the business of giving service away. For most, it is a business,
run to make a profit. This means they are competitors, and
look at the user, and each other, as a customer and a source
of revenue. Providers tussle as they compete, and consumers
tussle with providers to get the service they want at a low
How can we, as engineers, shape the economic tussle? In
fact, we have great power to shape this tussle, but first we
have to understand the rules that define it. A standard
business saying is that the drivers of investment are fear
and greed. Greed is easy to understand—it drove hundreds
of billions of dollars worth of investment in telecommunications over the last decade, much of which now sits at risk of
bankruptcy. But fear is more subtle. The vector of fear is
competition, which results when the consumer has choice.
The tussle among providers and consumers in a competitive
landscape is the most basic attribute of a marketplace. Most
economists of a “western” bent would argue that competition is good: it drives innovation, disciplines the market,
insures efficiency, and removes the need for intervention and
regulation of a market. To make competition viable, the
consumer in a market must have the ability to choose. So
our principle that one should design choice into mechanism
is the building block of competition.
Here are some specific examples, with implications for research and network design:
Value pricing
One of the standard ways to improve revenues is to find
ways to divide customers into classes based on their willingness to pay, and charge them accordingly—what economists
call value pricing. An example from another sector is the
“Saturday night stay” criterion for airline travel. It costs
the airline no more to carry a passenger if she does not stay
over Saturday night, but this restriction tends to separate
the business and pleasure traveler, which is useful because
the business traveler seems to have a greater willingness to
pay. Airlines impose Saturday night stay restrictions, and
consumers respond by buying multiple tickets, and using
only some of the segments of the flight. Airlines respond by
declaring this behavior unacceptable. And thus the tussle
As an example of similar behavior in the Internet, some
acceptable use policies for residential broadband access prohibit the operation of a server in the home. To run a server,
the customer is required to pay a higher “business” rate.
Customers who wish to sidestep this restriction can respond
by shifting to another provider, if there is one, or by tunneling to disguise the port numbers being used. The probable
outcome of this tussle depends strongly on whether one perceives competition as currently healthy in the Internet, or
eroding to dangerous levels.
• This discussion illustrates the observation that there
may be no such thing as value-neutral design. The design and deployment of tunnels (or other mechanisms
to mask what services are being used by a consumer)
shifts the balance of power from the producer to the
consumer. Given that value pricing is not a moral
wrong, should the consumers be aided in their quest
to bypass the controls of the producers? Those who
Provider lock-in from IP addressing
There is now considerable interest in the economics community in the nature of the Internet. Some of the seminal
papers are published in [9]. For an overview of the current
literature on Internet economics, see the Web site maintained by Mackie-Mason at
whole, because they allow each tussle to play out independently.
see the consumer as “the little guy” being abused by
the “big providers” will design such mechanisms, and
this is part of the tussle, not something that happens
outside the tussle. What mechanisms get designed,
and what standards get approved, are all part of the
Competitive wide area access
Today, the Internet system does not let the individual customer select his “long distance provider” the way the telephone system does. This is an example of designers failing
to appreciate a competitive tussle space.
At the time that equal access was being introduced into
the telephone system, there was a call for Internet routing to
support the same capability. The Internet designers deemed
this not necessary. They reasoned that there would be sufficient competition in the market because there were going
to be many ISP’s directly competing to serve the customer.
Letting the local provider enter into a wholesale arrangement to obtain wide area service seemed adequate, because
if one local provider made an unsatisfactory choice in wide
area provider, the customer could just switch to a new local
But this decision may be having undesirable consequences
today. It is possible that customers today would be much
more likely to see more service diversity, e.g. quality of
service support for applications, if there were more competition.
Residential broadband access
There is concern today that the advent of broadband residential access will be accompanied by a great reduction in
competition. Today there are almost 6000 dialup Internet
service providers. A pessimistic outcome five years in the
future is that the average residential customer will have two
choices—his telephone company and his cable company, because they control the wires. This loss of choice and competition is viewed with great alarm by many, who fear that it
may lead to higher prices and restrictions on what the user
may do, and there are many forces aligning to fight this
loss of competition. Some are regulatory, calling for laws to
mandate “open access”, to force the owners of the wires to
allow multiple ISPs to use them. Economists and regulators
hope that multiple providers will install their own cables, to
increase competition.3 However, in a tussle of competition,
one cannot compel a potential provider to invest and enter
a market.
Using the principles of this paper, one should speculate
on what sorts of investments are actually likely to be made,
and to think about what choice, and what tussle modularity, would improve the outcome of such an investment. One
investment option that is gaining momentum now is municipal deployment of fiber, because fiber installed by a neutral
party such as a municipality can be a platform for competitors to provide higher level services (e.g. phone, Internet
or television). This requires that the equipment lighting
the fiber support multiple service providers. Most of the
equipment made today is not “naturally open” in this way,
having been designed without consideration of this particular modularity boundary (or indeed with the specific goal of
confounding it).
• The Internet should support a mechanism for choice of
source routing that would permit a customer to control the path of his packets at the level of providers.
A design for such a system must include where these
user-selected routes come from or how they are constructed, how failures are managed, and how the user
knows that the traffic actually took the desired route.
The capability must also be approachable by a broad
class of users of varying sophistication. This is a very
complex design challenge,4 but could have a great influence.
This example illustrates another important point about
competition. One should be prepared to pay for what one
uses, or there is little incentive for a provider to offer it. Today, service providers do not like loose source, because ISPs
don’t receive any benefit when they carry traffic directed by
a source route. ISPs enter into business arrangement that
determine which traffic they agree to carry across which interfaces, and a source route has the effect of overriding these
arrangements. Why should they be enthusiastic about this?
Since source routes don’t work effectively today, researchers
propose even more indirect ways of getting around providerselected routing, such as exploiting hosts as intermediate
forwarding agents. (This kind of overlay network is a tool
in the tussle, certainly.) Another, perhaps simpler, approach
is to compensate the provider for carrying the packets. But
this idea tends to upset designers as well as customers, because they fear they will end up in an onerous “pay by the
byte” situation, which does not seem to have much market
• An important R&D project is to design and demonstrate a fiber-based residential access facility that supports competition in higher-level services. Technical
questions include whether sharing should be in the
time domain (packets) or color domain, how the fairness of sharing can be enforced and verified, an approach to fault isolation and other operational issues,
and how incremental upgrades can be done. This project is motivated both by the principle of “design for
choice”, and as well by recognition of new tussle boundaries.
Most of today’s “open access” proposals fail to balance
the interests of concerned parties because they are not modularized along tussle space boundaries. For example, the
capital costs and deployment pragmatics of broadband infrastructure differ greatly from those of operating mail and
web servers. This creates a natural boundary between the
two tussle spaces of broadband facilities provision and ISP
services. Proposals that implement open access at this modularity boundary are more likely to benefit the Internet as a
• The design for provider-level source routing must incorporate a recognition of the need for payment. There
In particular, today’s loose source routes, even if widely
implemented, would provide only a small portion of what is
For an analysis of issues in residential broadband access,
see [3].
trols.6 Issues of choice arise: who gets to pick which
firewall a user uses?
must be enough generality in the payment schemes
that the market can select an outcome that works for
all parties. (Remember, we are not designing the outcome, only the playing field for the tussle.)
• To prevent DoS attacks, protocols could be changed so
that end-nodes do not have to establish state or otherwise invest effort until they have verified that they
want to talk to the party initiating communication.
This concept is challenging, first because of a difficult balance between cost and function, and second,
because the idea of “first packet trust verification” is
at odds with the layered model of protocols, in which
transport establishes a connection before any higher
level information is exchanged.
• Overlay architectures should be evaluated for their
ability to isolate tussles and provide choice. A comparison is warranted between overlay architectures and
integrated global schemes to understand how each balances the relative control that providers and consumers
have, and whether economic distortion is greater in one
or the other.
Another tussle about firewalls is worth noting. Who gets
to set the policy in the firewall? The end user may certainly
have opinions, but a network administrator may as well.
Who is “in charge”? There is no single answer, and we better not think we are going to design it. All we can design is
the space for the tussle. But this illustrates the point about
visibility of decision-making. If a system administrator has
installed control rules in a firewall that affect an end user,
should that end user be able to download and examine these
rules? One way to help preserve the end to end character of
the Internet is to require that devices reveal if they impose
limitations on it. However, there is no obvious way to enforce this requirement, so it becomes a courtesy, not a real
Another dimension of trust is the fact that most users
don’t trust many of the parties they actually want to talk
to. We connect to web sites but are suspicious that they are
gathering information on us, stealing our credit cards, not
going to deliver what they promised, and so on. In this case,
the solution is more complex; we depend on third parties to
mediate and enhance the assurance that things are going
to go right. Credit card companies limit our liability to
$50, or sometimes nothing, in case of dispute. Public key
certificate agents provide us with certificates that assure us
we are talking to the party we think we are. Web sites assess
and report the reputation of other sites. The fact of these
third parties contrasts with our simple model of two-party
end-to-end communication among trusting parties. Each
individual interaction may be two-party end-to-end, but the
application design is not.
One of the most profound and irreversible changes in the
Internet is that by and large, many of the users don’t trust
each other. The users of the Internet no longer represent
a single community with common motivation and shared
trust. There are parties with adverse interests, and some
genuine “bad guys” out there. This implies that mechanisms
that regulate interaction on the basis of mutual trust should
be a fundamental part of the Internet of tomorrow.5
Most users would prefer to have nothing to do with the
bad guys. They would like protection from system penetration attacks, DoS attacks, and so on. This is a profound
tussle, between people who want to be left alone, and people who want to bother them. Since host security today is
of variable and mostly poor quality, this desire for protection leads to firewalls. Firewalls change the Internet from a
system with transparent packet carriage between all points
(what goes in comes out), to a “that which is not permitted
is forbidden” network. This is a total reversal of the Internet philosophy, but pure transparency is not what most
users long for. For over ten years, Internet purists have been
bemoaning the fact that firewalls inhibit innovation and the
introduction of new applications (fifteen years ago they were
called “mail gateways”), but firewalls have not gone away.
The principle of “design for choice” would imply that users
should be able to choose with whom they interact, and users
should be able to choose the level of transparency they offer
to other users. The principle of “tussle isolation” suggests
that these mechanisms should not be overloaded on to any
other mechanism, but should be separated. Further, one
should consider if within the broad topic of trust, there are
separable issues.
The first topic is control over which parties are willing to
exchange packets with each other.
• An important engineering principle for future applications is that there should be explicit ability to select
what third parties are used to mediate an interaction,
and to act as an agent for the end-user in improving
his trust in the operation. The parties must be able
to choose, so they can select third parties that they
• In the abstract, there is a technical question as to
whether each end-node can implement sufficient trustrelated controls within itself, or whether delegation
of this control to a remote point inside the network
is required—a “trust-aware firewall”. As a practical
matter, the market calls for firewalls. Firewalls of the
future must be designed so that they apply constraints
based on who is communicating, as well as what protocols are being run and where in the network the parties
are. Such a device would imply the design of new protocols and interfaces, to allow the end node and the
control point to communicate about the desired con-
Another space in which trust is eroding is that users less
and less trust the software they have to run. They suspect
The IETF is working on such standards, e.g. the MIDCOM
working group.
An interesting debate relevant to this topic centers on the
IETF proposal to charter the Open Pluggable Edge Services
(OPES) working group, and the IAB deliberation on policy
concerns. The IAB has focused on issues of whether one end
or both have to concur with the insertion of an intermediate
node in the communication, and what tools the user should
have to detect and recover from a faulty node.
A thoughtful analysis of trust that has shaped our thinking
is provided by [10].
The openness to innovation—to new applications and new
uses—has perhaps been the most critical success factor for
the Internet. But openness is not an unalloyed virtue for
service providers. Openness often equates to competition,
which creates the fear factor that demands costly investment
and drives profits to a minimum. Many telephone company
executives remember the good old monopoly days, with a
comfortable regulated rate of return and no fear. And many
current ISPs may long for a return to those less open, high
margin days, if they could only figure out how to get there.
The keys are closed or proprietary interfaces and vertical
Motivations concerning open vs. proprietary systems have
much to do with economics. Economists have studied the
motivation of providers with various degrees of market power
to choose open or proprietary interfaces; see [6]. Industry
understands that interfaces, or lack thereof, can shape a
market.9 There is probably a whole paper on the tussles
surrounding open vs. closed systems. However, as a starting point, the first exercise should be to speculate about
whether these various openness tussles can be modularized
and disentangled, and what this means for mechanism design.
Vertical integration—the bundling together of infrastructure and higher-level services—requires the removal of certain forms of openness. The user may be constrained to
use only certain providers of content, or to pay to run certain protocols, and so on. However, vertical integration has
nothing to do with a desire to block innovation. Even in a
market with a high degree of vertical integration, innovation
that brings new value to the customer is likely to benefit all
parties. So it would be wise to separate the tussle of vertical
integration, about which many feel great passion, from the
desire to sustain innovation.
The technical characteristic of the network that has fostered innovation is transparent packet carriage—the ability
to deploy a new protocol without having to modify the inside
of the network. But transparency is not the same thing as
openness, though they are related. With this brief motivation, we consider some old design principles of the Internet,
including the principle that is usually equated with transparency, the end to end arguments.
their operating system and browser of gathering information
on them and passing it on without their knowledge, or turning them in for software license violations. There are web
sites that claim to look at the outgoing data stream from
the user’s machine and detect and remove any information
that is leaking out.
• This problem may best be dealt with using non-technical means—regulation, public opinion and so on. Just
because a problem manifests in a technical space, it
does not mean it has to be solved there. But it is an
interesting exercise to consider whether there are technical means to protect a user from software running on
their own machine. The history of mandatory security
controls and security kernels suggests that this problem is thorny.
The role of identity
One obvious point about trust is that if communication is
to be mediated based on trust, then as a preliminary step,
parties must be able to know to whom they are talking.
Otherwise, one has little basis for judging how much to trust
One could take this as a call for the imposition of a global
namespace of Internet users, with attached trust assessments. We believe this is a bad idea. It is hard to imagine
a global system that is really trustworthy. More importantly, there are lots of ways that parties choose to identify
themselves to each other, many of which will be private to
the parties, based on role rather than individual name, etc.
What is needed is a framework that translates these diverse
ways into lower level network actions that control access.
This implies a framework for talking about identity, not a
single identity scheme. We suggest that such a framework
could usefully share and arbitrate information across many
layers of the protocol stack.
The need to know to whom we are talking will challenge a
current precept of the Internet, which is that it is permissible
to be anonymous on the Internet. There is a fundamental
tussle between the ideas of anonymous action, and the idea
that in a society where “that which is not forbidden is permitted”, one can be held accountable for ones actions. A
possible outcome of this tension is that while it will be possible to act anonymously, many people will choose not to
communicate with you if you do, or will attempt to limit
what you do.8 A compromise outcome of this tussle might
be that if you are trying to act in an anonymous way, it
should be hard to disguise this fact. This illustrates the
observation that one must think about whether the consequences of choice are visible, or can be hidden.
The future of the end to end arguments
One of the most respected and cited of the Internet design
principles is the end to end arguments, which state that
mechanism should not be placed in the network if it can be
placed at the end node, and that the core of the network
should provide a general service, not one that is tailored to
a specific application [11]. There are two general dimensions
to the arguments: innovation and reliability.
The tussles of openness
One of the most profound fears for the Internet today is
that it will lose its “open” qualities: the openness to innovation that permits a new application to be deployed, the
openness of access that allows a user to point their Web
browser at any content they please, the openness that allows a user to select the servers and services that best meet
their needs.
While technical network designers may not think about
open interfaces as a tool to drive market structure, industrial players understand this fully. When then Senator Gore
announced his vision for a National Information Infrastructure (NII) in the early 1990s, at least two organizations produced requirement documents for the “critical interfaces”
that would permit the NII to have a suitable structure [4,
5, 2].
An analog is the current situation with Caller ID, where
a sender can block the caller’s information, but the receiver
can refuse to accept calls from a sender that does.
Innovation: If the core of the network has been tailored to
one specific application, this may inhibit the deployment of
other applications. If the core of the network must be modified to deploy a new application, this puts a very high hurdle
in front of any unproven idea, and almost by definition, a
new idea is unproven.
Reliability and robustness: If bits of applications are “in
the network”, this increases the number of points of failure
that can disable the application. The more simple the core
of the network, the more reliable it is likely to be.
The simplest application of the end to end arguments produces a network that is transparent : packets go in, and they
come out, and that is all that happens in the network. This
simple idea was very powerful in the early days of the Internet, but there is much fear that it seems to be eroding, for
many of the reasons discussed above:
new applications must, almost of necessity, launch incrementally, they most benefit from the transparent simplicity that
the end to end arguments fostered. By the principle of isolation of tussle, any barriers that are put into the network as
a result of the desire to control mature applications or issues
of trust should not prevent parties that want transparency
from getting it.
Failures of transparency will occur—design what
happens then. Today, when an IP address is unreachable,
there is little in the way of helpful information about why.
A sophisticated user can run traceroute, but today’s normal
user just gets frustrated. Tools for fault isolation and error
reporting would help – the hard challenge is not so much to
find the fault but to report the problem to the right person
in the right language. That person may be someone who
can fix the problem, or someone who can decide to choose
a different path or provider – fault reporting is as much a
tool of tussle management as it is a tool of technical repair. Of course, some devices that impair transparency may
intentionally give no error information or even reveal their
presence, and that must be taken into account in design of
diagnostic tools.10
Peeking is irresistible. If there is information visible
in the packet, there is no way to keep an intermediate node
from looking at it. So the ultimate defense of the end to
end mode is end to end encryption. In most discussions of
the need for encryption, the putative threat is someone who
wants to steal your information or modify it. The ISP as
a “threat” is not the normal assumption. But if the ISP is
trying to control or modify what you are doing, then the
ISP is the issue.
Of course, encrypting the data stream has drawbacks.
One is that the actions of the ISP might actually be making
things better. They might be offering performance improvements or other benefits that the end user actually wants.
But this situation is not an issue; if the user has control over
whether the data is encrypted or not, the user can decide
if the ISP actions are a benefit or a hindrance. The other
drawback is that encrypting the stream might just be the
first step in an escalating tussle between the end user and
the network provider, in which the response of the provider
is to refuse to carry encrypted data. It is probably not the
case that a commercial ISP would escalate to this level. In
the U.S., competition would probably discipline a provider
that tried to do this. But a conservative government with
a state-run monopoly ISP might. And in that case, policy
will probably trump technology in any case. Then the advantage of having the encrypted mode is that it would force
the government to be explicit about what their policy was.
Forcing the choice to be public and visible is about all that
technology can do to moderate this situation.11
Note that in a multi-way application, where third parties
are involved to insure the validity of the transaction, the
meaning of “end to end” gets more complex, and so does
the proper use of encryption.
• The lost of trust calls for less transparency, not more,
and we get firewalls.
• The desire for control by the ISP calls for less transparency, and we get application filtering, connection
redirection, and so on.
• The desire of third parties to observe a data flow (e.g.
wiretap) calls for data capture sites in the network.
• The desire to improve important applications (e.g. the
Web), leads to the deployment of caches, mirror sites,
kludges to the DNS and so on.
This is a lot of mechanism, a large potential loss of transparency, and an increasing focus on improving existing applications at the expense of new ones. So what is the future
of the end to end arguments? We argue that the end to
end arguments are still valid and powerful, but need a more
complex articulation in today’s world. The discussion to
this point gives us some guidance.
Evolution and “enhancement” of existing, mature
applications is inevitable. As applications become popular, lots of players, including application providers and the
ISPs, will want to get involved in them, whether as a move
toward vertical integration, enhancement of performance or
reliability, or some other reason. This will almost certainly
lead to increased complexity, perhaps decreased reliability
or predictability, and perhaps an evolution of the overall
application away from the original vision. We should not
imagine that anyone can do much about this. If we design
applications so that the user can control what features “in
the network” are invoked, we may have done as much as we
The most we can do to protect maturing applications is to bias the tussle. If application designers want
to preserve choice and end user empowerment, they should
be given advice about how to design applications to achieve
this goal. This observation suggests that we should generate “application design guidelines” that would help designers
avoid pitfalls, and deal with the tussles of success.
Keeping the net open and transparent for new applications is the most important goal. Innovation and
the launch of new applications is the engine that has driven
the growth of the Internet and the generation of new value.
So barriers to new applications are much more destructive
than network-based support of proven applications. Since
See the footnote above on the deliberations by the IAB on
the charter for the OPES working group.
The next step in this sort of escalation is steganography—
the hiding of information inside some other form of data.
It is a signal of a coming tussle that this topic is receiving
attention right now.
Separation of policy and mechanism
Another design principle of great age and uncertain origin12 is that technologists should design policy-free mechanism, and allow those who use the system (whether literal
“users”, administrators, etc) to adjust the mechanisms to
match their specific needs. This paper challenges this principle as perhaps being too simplistic. True value-neutral
design is, at best, extremely difficult. Mechanism defines
the range of “policies” that can be invoked, which is another way of saying that mechanism bounds the range of
choice. So in principle there is no pure separation of policy
from mechanism.
However, this does not negate the principle. The chief advantage of attempting to separate mechanism and policy is
to isolate some regions of the system from tussle. Even if the
attempt is not completely successful, these isolation regions
can serve to separate different tussles from each other, and
can serve as technological ‘fixed points’ that allow different
tussles to play out at different speeds.
This more complex interpretation of old design principles,
and the introduction of new principles, needs to be seen in
terms of system synthesis. How can we, as designers, build
systems with desired characteristics and improve the chances
that they come out the way we want? If we try to design
a system that is open, for example, which means we will
encounter the tussles surrounding vertical integration and
capture of value in exchange for investment, how can we
One can learn from the past. To some of us in the research community, a real frustration of the last few years is
the failure of explicit QoS to emerge as an open end-to-end
service. This follows on the failure of multicast to emerge
as an open end-to-end service. It is instructive to do a postmortem on these failures.13 Here is one hypothesis. For the
ISPs to deploy QoS, they would have to spend money to upgrade routers and for management and operations. So there
is a real cost. There is no guarantee of increased revenues.
Why risk investment in this case? If the consumer could
exercise effective competitive pressure in ISP selection, fear
and greed might have driven ISPs to invest, but the competitive pressures were not sufficient. On the other hand, if
ISPs use the new QoS mechanisms in a closed way, rather
than an open way, they greatly enhance revenue opportunities. Thus, for example, if they deploy QoS mechanisms
but only turn them on for applications that they sell, they
reduce the open nature of the Internet and create opportunities for vertical integration. If Internet Telephony requires
QoS to work, and they only turn on QoS for their version
of Internet Telephony, then they can price it at monopoly
One can thus see the failure of QoS deployment as a failure first to design any value-transfer mechanism to give the
providers the possibility of being rewarded for making the
investment (greed), and second, a failure to couple the design to a mechanism whereby the user can exercise choice
to select the provider who offered the service (competitive
fear). The argument about choice here is actually subtle.
The user had the power to choose the level of QoS needed—
that could be expressed in the ToS bits. What was missing
was routing, to allow the user to favor one ISP over another
if that ISP honored the bits.
• Perhaps the most challenging intellectual puzzle in this
design space is to discover parts of mechanism that
really can be divorced from policy — which, in other
words, actually are value-neutral.
One value (or bias) that is shared by many people is user
empowerment. This is the preference that the user, rather
than the service provider or the software provider, be able
to pick what applications to run, what servers and services
to use, and so on. User empowerment, to many, is a basic
Internet principle, but for this paper, it is the manifestation
of the right to choose—to drive competition, and thus drive
One could argue that user empowerment is a bias, of the
“David and Goliath” sort—a bias imposed on the tussle between the little guy and the provider, who is seen as “big and
bad”. This view would suggest that to the extent one tries
to be value-neutral in the design of mechanism, one should
not favor user empowerment. One could also argue that the
fundamental design goal of the Internet is to hook computers together, and since computers are used for unpredictable
and evolving purposes, making sure that the users are not
constrained in what they can do is doing nothing more that
preserving the core design tenet of the Internet. In this
context, user empowerment is a basic building block, and
should be embedded into all mechanism whenever possible.
This paper suggests that the latter view is the defensible
one, because choice is a basic tool to deal with tussle.
The recognition of tussle as a fundamental behavior does
give one further hint at how to try to separate mechanism
from policy. If one can find spaces where tussle are unlikely,
then (as noted above) the interfaces and mechanisms can
be simpler. If one can truly separate tussles, then one can
do a better job of matching mechanism to problem. So the
instruction to “separate mechanism from policy” is not incorrect, but just requires careful thought to carry out as best
one can.
• Anyone who designs a new enhancement for the Internet should analyze the tussles that it will trigger, and
the tussles in the surrounding context, and consider
how they can be managed to ensure that the enhancement succeeds. As noted above, a powerful force is
the tussle of competition. Protocol design, by creating
opportunities for competition, can impose a direction
on evolution.
As the Internet evolves to become a full component of
society, the person most likely to be dismayed is the fabled
cypherpunk. [7] summarizes the cypherpunk view of privacy
as follows: “[T]he cypherpunk’s credo can be roughly paraphrased as ‘privacy through technology, not through leg13
The case study of the failure to deploy multicast is left as
an exercise for the reader.
An early articulation of the principle can be found in [8].
[2] Computer Science and Telecommunications
Board, National Research Council. Realizing the
information future: The Internet and beyond, June
[3] Computer Science and Telecommunications
Board, National Research Council. Broadband:
Bringing home the bits, January 2002.
[4] Computer Systems Policy Project. Perspectives
on the national information infrastructure: Ensuring
interoperability, 1994.
[5] Cross-Industry Working Team. An architectural
framework for the national information infrastructure,
[6] Economides, N. The economics of networks.
International Journal of Industrial Organization 14, 6
(1996), 670–699.
[7] Goldberg, I., Wagner, D., and Brewer, E.
Privacy-enhancing technologies for the internet. In
Proceedings of IEEE COMPCON 97 (1997),
pp. 103–109.
[8] Levin, R., Cohen, E. S., Corwin, W. M.,
Pollack, F. J., and Wulf, W. A.
Policy/mechanism separation in HYDRA. In
Symposium on Operating Systems Principles (1975),
pp. 132–140.
[9] McKnight, L., and Bailey, J., Eds. Internet
Economics. MIT Press, 1997.
[10] Nissenbaum, H. Securing trust online: Wisdom or
oxymoron. Boston University Law Review (2001).
Available as http:
[11] Saltzer, J., Reed, D., and Clark, D. D.
End-to-end arguments in system design. ACM
Transactions on Computer Systems 2, 4 (Nov. 1984).
islation.’ If we can guarantee privacy protection through
the laws of mathematics rather than the laws of men and
whims of bureaucrats, then we will have made an important
contribution to society. It is this vision which guides and
motivates our approach to Internet privacy.” Our position
is that the laws of men and the so-called whims of bureaucrats are part of the fabric of society, like it or not. They
are some of the building blocks of tussle, and must be accepted as such. We, as technical designers, should not try
to deny the reality of the tussle, but instead recognize our
power to shape it. Once we do so, we acquire a new set of
hard, technical problems to solve, and this is a challenge we
should step up to willingly.
The authors gratefully acknowledge essential and ongoing
discussions with members of the NewArch project,14 particularly Mark Handley, Noel Chiappa, Ted Faber, and Aaron
Falk. Sharon Gillette, Jean Camp and the Sigcomm reviewers provided welcome comments and feedback, shaping
our future work as well as this paper. Sally Floyd provided
invaluable encouragement at a well chosen moment. Our
sincere thanks to all.
[1] Blumenthal, M. S., and Clark, D. D. Rethinking
the design of the Internet: The end to end arguments
vs. the brave new world. ACM Transactions on
Internet Technology 1, 1 (August 2001). Version
appeared in Communications Policy in Transition:
The Internet and Beyond, B. Compaine and S.
Greenstein, eds. MIT Press, Sept. 2001.
Без категории
Размер файла
196 Кб
633059, 964725
Пожаловаться на содержимое документа